European Journal of Clinical Nutrition (2015), 1–7 © 2015 Macmillan Publishers Limited All rights reserved 0954-3007/15 www.nature.com/ejcn
ORIGINAL ARTICLE
Fish consumption and all-cause mortality: a meta-analysis of cohort studies L-G Zhao1,2, J-W Sun1,2, Y Yang1,2, X Ma1,2, Y-Y Wang1,2 and Y-B Xiang1,2 BACKGROUND/OBJECTIVES: Although fish consumption may have an influence on specific mortality of major chronic diseases, the relationship between fish consumption and all-cause mortality remains inconsistent. SUBJECTS/METHODS: We performed a systematic search of publications using PubMed and Web of science up to 31 December 2014. Summary relative risk (RR) for the highest versus lowest category of fish consumption on risk of all-cause mortality was calculated by using a random effects model. Potential nonlinear relation was tested by modeling fish intake using restricted cubic splines with three knots at fixed percentiles of the distribution. RESULTS: Twelve prospective cohort studies with 672 389 participants and 57 641 deaths were included in this meta-analysis. Compared with the lowest category, the highest category of fish intake was associated with about a 6% significantly lower risk of all-cause mortality (RR = 0.94, 95% confidence interval (CI): 0.90, 0.98; I2 = 39.1%, P = 0.06). The dose–response analysis indicated a nonlinear relationship between fish consumption and all-cause mortality. Compared with never consumers, consumption of 60 g of fish per day was associated with a 12% reduction (RR = 0.88, 95% CI: 0.83, 0.93) in risk of total death. CONCLUSIONS: These results imply that fish consumption was associated with a reduced risk of all-cause mortality. European Journal of Clinical Nutrition advance online publication, 13 May 2015; doi:10.1038/ejcn.2015.72
INTRODUCTION Noncommunicable diseases, such as cancers, cardiovascular diseases and type 2 diabetes, are responsible for about two-thirds of deaths worldwide.1 Therefore, it is urgent to develop efficient strategies that can prevent noncommunicable diseases by reducing their major risk factors and recommending healthy lifestyles, including a well-balanced diet. Fish, as a widely consumed food of human diet, has many health benefits.2 A growing body of evidence indicates that high consumption of fish may decrease the risk of chronic diseases like coronary heart disease (CHD),3–5 stroke,6 type 2 diabetes7 and certain cancers8–12 and, consequently, also for all-cause mortality.13 A systematic review of 17 cohort studies showed that fish consumption had significantly beneficial effects on the prevention of CHD mortality.14 Recently, the Vitamins and Lifestyle Study (VITAL Study) reported that higher intake of fish was significantly associated with a lower risk of all-cause mortality.15 Fish is the most common dietary source of long-chain n-3 polyunsaturated fatty acids (n-3 PUFAs), which has been demonstrated to have antiatherosclerotic and antithrombotic effects.16,17 However, fish types, cooking methods, toxic metals and other environmental contaminants in fish may have different roles on human health.18 In addition, evidence from published epidemiological studies did not show consistent results between fish consumption and all-cause mortality.13,15,19–28 Hence, we conducted a meta-analysis to assess this relationship quantitatively. 1
MATERIALS AND METHODS Search strategy We performed a systematic search in PubMed (www.ncbi.nlm.nih.gov/ pubmed/) and Web of Science (www.webofknowledge.com/) for studies published before 31 December 2014. We used the search terms as ‘fish intake’, ‘fish consumption’ in combination with ‘mortality’, ‘death’ or ‘fatal’. Furthermore, we manually reviewed the reference lists of identified papers, previous reviews and relevant meta-analyses to identify any studies that were not found from the preliminary literature searches. Two investigators (L-GZ and J-WS) independently conducted the literature search. The study was eligible for inclusion if it had the following: (1) had a cohort design based on general healthy population and published in English, (2) used fish consumption as the exposure of interest, (3) identified total death or all-cause mortality as the outcome of interest and (4) reported the relative risk (RR) or hazard risk (HR) (highest versus lowest) and the corresponding 95% confidence interval (CI).
Data extraction From each eligible study, we extracted the first author’s name, study population and region, study period, number of participants, gender, exposure measurement, covariates adjusted RR or HR and the corresponding 95% CI, as well as numbers of cases and person-years or persons for each category of fish consumption. We used the risk estimates that derived from the multivariable-adjusted model that did not include the potential intermediates (such as, blood glucose, triglyceride or inflammatory biomarkers) on the causal pathway from exposure to disease in original studies, because control for an intermediate variable would bring overadjustment bias.29 Data extraction was implemented independently by two authors. Any inconsistency was solved by the third author (YY). In order to perform a dose–response analysis, when fish consumption was measured as ‘servings per week’, we converted this into ‘gram per day’
State Key Laboratory of Oncogene and Related Genes, Shanghai Cancer Institute, Renji Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, China and Department of Epidemiology, Shanghai Cancer Institute, Renji Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, China. Correspondence: Professor Y-B Xiang, Department of Epidemiology, Shanghai Cancer Institute, Renji Hospital, Shanghai Jiaotong University School of Medicine, No. 25, Lane 2200, Xie Tu Road, Shanghai 200032, China. E-mail: [email protected] Received 15 August 2014; revised 16 March 2015; accepted 6 April 2015 2
Fish consumption and all-cause mortality L-G Zhao et al
2 as a standard measure of fish intake according to the information provided by the identified study.24 For each category, we assigned the median value of each fish category. If the upper bound in the highest category was not available, we assumed that it had the same amplitude as the preceding one. If the distribution of cases or person-years in the study was not reported, we estimated the distribution using the methods described by Aune et al.30
Records identified through PubMed (n=982) and Web of Science (n=2445) Records with duplicate titles /abstracts were excluded (n=822). Records after duplicates removed (n=2605) Records were excluded based on screening of titles and/or abstracts using general criteria (n=2564)
Statistical analysis We conducted two types of meta-analyses. First, we combined the RRs for the highest versus lowest category of fish consumption on risk of all-cause mortality by using the DerSimonian and Laird random-effects model, which incorporated both within- and between-study variability.31,32 Second, for the dose–response meta-analysis, the method proposed by Greenland and Longnecker33 and Orsini et al.34 was used to calculate the trend from the correlated log-relative risks across categories of fish consumption. In order to avoid bias estimates, we centered each study to the baseline reference of fish intake.35 Potential nonlinear relation was tested by modeling fish intake using restricted cubic splines with three knots at fixed percentiles (10, 50 and 90%) of the distribution.32,34 A P-value for nonlinearity of the meta-analysis was evaluated by testing the null hypothesis that the coefficient of the second spline was equal to zero.36 Heterogeneity among studies was evaluated by using the Q and I2 statistics. I2 took values between 0 and 100% and there may be important heterogeneity when I2450% or P for Q statistic o0.10.37 In a sensitivity analysis, we sequentially omitted one study at a time from the metaanalysis to examine whether the main results were influenced by a particular study. In order to test the robustness of the combined estimates, separate meta-analysis was performed stratified by gender (men or women), region (USA, Asia or Europe), methods of dietary assessment (validated food frequency questionnaire (FFQ) or others), sample size (using 40 000 participants as the cut point), follow-up duration (using 12 years as the cut point), year of publication (before 2008 or after 2008) and important confounding adjustments (yes or no) for education, body mass index, physical activity, intake of fruit and vegetables, red meat and total energy. Small study bias (for example, publication bias) was assessed through visual inspection of funnel plots and using Egger’s38 and Begg’s tests.39 P o 0.10 was deemed to possess publication bias. All statistical analyses were conducted using STATA software (version 12.0; StataCorp, College Station, TX, USA). P-values with two side of o0.05 were considered statistically significant if not specified.
RESULTS Literature search and study characteristics We obtained 2605 articles after removal of duplicates from our preliminary search. After screening the title and abstract, 2564 records were excluded using the general criteria, such as review, animal research and retrospective study. Forty articles were potentially related to our issue for further scrutiny, of which 29 articles were excluded according to the inclusion criteria. With one study19 obtained from checking reference lists, thus twelve studies13,15,19–28 were included in the meta-analysis (Figure 1). Takata et al.20 reported the results of Shanghai Women Health Study (SWHS) and the Shanghai Men Health Study (SMHS), whereas the SWHS was updated by a pooled cohort study13 including eight cohorts in Asia. Then, we only abstracted the data of the SMHS from this study20 in meta-analysis of highest versus lowest category. Of the twelve identified studies (Table 1), the median follow-up time was 12 years. Combined, these studies included 672 389 cohort members and 57 641 deaths. Six studies15,19,24,26–28 were conducted in the United States, four13,20,22,25 in Asia and two21,23 in Europe. Most studies13,15,19–22,24–27 used a structured FFQ to estimate fish intake. All studies adjusted for age and gender as potential confounders (if applicable). Only one study23 failed to adjust for smoking status and alcohol drinking. European Journal of Clinical Nutrition (2015) 1 – 7
Records obtained from titles/abstracts screening (n=41)
Records obtained from full-text screening (n=11)
Records were excluded as: (1) No RRs/HRs or 95%CIs for highest/lowest were reported (n=9); (2) Dietary pattern rich in fish or sea food as exposure is reported (n=11); (3) The outcome is specific disease mortality (n=5); (4) The base population are patients (n=5). Records obtained from checking reference lists of retrieved articles (n=1).
Studies included in meta-analysis (n=12) (1) For highest/lowest analysis (n=12); (2) For dose-response analysis (n=7).
Figure 1. Flow diagram for selection of studies in meta-analysis of fish consumption and all-cause mortality. RR, relative risk; HR, hazard ratio; CI, confidence interval.
Highest versus lowest category Overall, analysis of 12 prospective studies significantly showed a 6% reduction in risk of all-cause mortality with high intake of fish (RR = 0.94, 95% CI: 0.90, 0.98) with moderate evidence of betweenstudy heterogeneity (I2 = 39.1%, P = 0.06) using a random-effects model (Figure 2). Visual inspection of Begg’s test and Egger’s test suggested that there was no publication bias (Begg’s test: P = 0.26; Egger’s test: P = 0.14). Sensitivity analysis and subgroup analysis In sensitivity analysis, we recalculated the combined results by sequentially excluding each study to examine the influence of individual study on the pooled estimate. The risk estimates ranged from 0.92 (95% CI: 0.89, 0.95; I2 = 0.0%, P = 0.462), after excluding Olsen,21 to 0.95 (95% CI: 0.90, 0.99; I2 = 37.4%, P = 0.072), after excluding Bell.15 None of the studies considerably affected the summary results. Subgroup analysis was performed across a number of key study characteristics (Table 2). The pooled RRs did not materially differ between genders. A lower risk was observed in the United States, whereas no association was found in Europe. There was a significantly inverse association in studies published after 2008. Stratifications by other characteristics did not have a substantial impact on the main results. Dose–response meta-analysis Although we had contacted with the authors for more detailed information, five studies were excluded from dose–response analysis because only two categories of fish consumption were considered,21 or no information was provided regarding the amount of fish intake in each category.13,15,19,26 © 2015 Macmillan Publishers Limited
© 2015 Macmillan Publishers Limited
Japan
UK
Yamagishi22
Ness23
6
7
China
Takata20
4
Denmark
Asia
Lee13
3
Olsen21
USA
Kappeler19
2
5
USA
Bell15
1
37
12.7
12
5.6
6.6–15.6
22
5
Cohort size
British
Japanese
Danes
Chinese
NA
7008
4126 (2383 men and 1743 women)
2170
24 283 (14 326 men and 9957 women)
3683 (1908 men and 1775 women)
3051
No. of deaths
4028 (1995 1010 men and 2033 women)
57 972 (22 881 men and 35 091 women
50 290 (23 274 men and 27 016 women)
73 159
296 721 (112 310 men and 184 411 women)
Mixed: main 17 611 non-hispanic white (76.5%)
Mixed: main 70 495 white (93.3%)
Follow-up Populations years
The food records
FFQ with 33 items
FFQ with 192 items
FFQ
FFQ
FFQ with 81 items
FFQ with 120 items
Dietary assessment method
General characteristics of prospective studies of fish consumption and all-cause mortality
ID First author Country (reference no.)
Table 1.
44.5 versus 1.8 g per day
Men: 86 versus o19 g per day. Women: ⩾ 84 versus o19 g per day
Men: ⩾ 41 versus o41 g per day. Women: ⩾ 35 versus o35 g per day
121.9 versus 11.1 g per day
Both
Both
Both
Men
Both
Age, sex, energy, childhood family food expenditure, father’s social class, district of residence as a child, period of birth, season when studied as a child and Townsend score for current address or place of death.
Age, gender, energy, history of hypertension and diabetes mellitus, smoking status, alcohol consumption, BMI, mental stress, walking, sports, education levels, total energy, and dietary intakes of cholesterol, saturated and n-6 polyunsaturated fatty acids, vegetables and fruit.
Age, time under study, smoking status, smoking duration, current tobacco consumption, time since cessation, alcohol intake, school education, participation in sports, time spent on sports per week, BMI, intake of red meat, intake of processed meat intake, total energy intake and other dietary components.
Age at baseline, total energy intake, income, occupation, education, comorbidity index, physical activity level, red meat intake, poultry intake, total vegetable intake, total fruit intake, smoking history and alcohol consumption.
Age, BMI, education, smoking habit, rural/urban residence, alcohol intake, fruit and vegetable intake and total energy intake.
Age, race, sex, cigarette smoking, alcohol consumption, physical activity, socioeconomic status, BMI, marital status, fruit and vegetables intake, history of hypertension, diabetes, hypercholesterolemia, use of aspirin and ibuprofen, use of mineral and vitamin supplements, family history of diabetes or hypercholesterolemia; hormone replacement therapy and oral contraceptive use (in women).
49 versus 0 times per Both month
Q4 versus Q1: mean = 36.6 6 ± 37.2 g per day
Age, sex, race/ethnicity, marital status, education, total energy intake, BMI at age 45 years, average alcohol intake at age 45 years, average physical activity in the 10 years before baseline, self-rated health, mammogram in the last 2 years, prostate-specific antigen test in the last 2 years, sigmoidoscopy in the last 10 years, current use of cholesterol-lowering medication, aspirin use in the past 10 years, use of nonaspirin nonsteroidal anti-inflammatory drugs in the past 10 years, smoking, morbidity score, percentage of calories derived from trans fat, percentage of calories derived from saturated fat, number of servings per day of fruits, number of servings per day of vegetables, years of estrogen therapy, years of estrogen+progestin therapy, age at menopause, age at death of father, and age at death of mother.
Both
442 versus 0 times per year
Adjustment
Sex
Quantity
Fish consumption and all-cause mortality L-G Zhao et al
3
European Journal of Clinical Nutrition (2015) 1 – 7
(Continued )
European Journal of Clinical Nutrition (2015) 1 – 7
USA
USA
11 Albert27
12 Daviglus28
25.8
11
18.8
6.9
10.5
29 079 (13 355 men and 15 724 women)
41 836
Cohort size
20 551
Chicago men 1822
US male physicians
Mixed: main 8825 white (84.1%)
Japanese
USA postmenopausal women
Follow-up Populations years
Abbreviations: BMI, body mass index; FFQ, food frequency questionnaire.
USA
Japan
Nagata25
9
10 Gillum26
USA
Folsom24
8
ID First author Country (reference no.)
Table 1.
1042
1652
2901 (1513 men and 1388 women)
2062 (1163 men and 899 women)
4653
No. of deaths
Men
⩾ 5 times per week versus ⩽ 1 time per month
Men
Baseline age, smoking, history of diabetes, education, high school graduate, systolic blood pressure, serum cholesterol concentration, BMI, alcohol intake and physical activity.
Both
41 times per week versus never
Baseline age, education, religion, systolic pressure, serum cholesterol, number of cigarettes smoked per day, BMI, presence or absence of diabetes, presence or absence of electrocardiographic abnormalities, daily intake of energy, cholesterol, saturated, monounsaturated, fatty acids, total protein, carbohydrate, alcohol, iron, thiamine, riboflavin, niacin, vitamin C, beta carotene and retinol.
Age, aspirin and beta carotene treatment assignment, evidence of cardiovascular disease before 12-month questionnaire, BMI, smoking status, history of diabetes, history of hypertension, history of hypercholesterolemia, alcohol consumption, vigorous exercise, and vitamin E, vitamin C, multivitamin use, other dietary factors (red meat, vegetables, fruits, dairy, chicken or turkey and fried foods).
For men: age, total energy, marital status, BMI, smoking status, alcohol intake, coffee intake, exercise, and history of hypertension and diabetes mellitus. For women: age, total energy, marital status, years of education, BMI, smoking status, alcohol intake, age at menarche, menopausal status, exercise and history of diabetes mellitus.
Men: 157.8 versus Both 46.2 g per day. Women: 122.4 versus 36.6 g per day
Women Age, energy intake, educational level, physical activity level, alcohol consumption, smoking status, pack-years of cigarette smoking, age at first livebirth, estrogen use, vitamin use, BMI, waist/hip ratio, diabetes, hypertension, intake of whole grains, fruit and vegetables, red meat, cholesterol and saturated fat.
42.5 versus o0.5 serving per day
Adjustment
Sex
Quantity
Standard ⩾35 versus 0 g questionnaires per day
FFQ
3-month FFQ
FFQ with 169 items
FFQ with 127 items
Dietary assessment method
Fish consumption and all-cause mortality L-G Zhao et al
4
© 2015 Macmillan Publishers Limited
Fish consumption and all-cause mortality L-G Zhao et al
5 study
year
sex
quantity
Bell et al.
2014
Both
>42 times/y vs. 0 times/y
0.86 (0.76, 0.98)
Kappeler et al.
2013
Both
>9 times/m vs. 0 times/m
0.93 (0.78, 1.11)
Lee et al.
2013
Men
Q4 vs. Q1: mean=36.6 g/d
1.05 (0.95, 1.16)
Lee et al.
2013
Women Q4 vs. Q1: mean=36.6 g/d
0.91 (0.85, 0.97)
Takata et al.
2013
Men
121.9 g/d vs. 11.1 g/d
0.86 (0.74, 1.00)
Olsen et al.
2011
Men
>=41 vs. =35 vs. =86 vs. =84 vs. 2.5 vs. 1 times/w vs. never
0.89 (0.73, 1.09)
Gillum et al.
2000
Women >1 times/w vs. never
0.88 (0.71, 1.09)
Albert et al.
1998
Men
>=5 times/w vs. =35 g/d vs. 0 g/d
0.85 (0.64, 1.10)
RR (95% CI)
Overall (I-squared = 39.1%, p = 0.055)
0.94 (0.90, 0.98)
NOTE: Weights are from random effects analysis
0.5
0.6
0.7
0.8
0.9
1
1.2
Figure 2. Relative risk of all-cause mortality for highest versus lowest category of fish intake. Overall relative risk calculated with random effects model.
We observed a significantly nonlinear relationship (P = 0.020) between fish consumption and all-cause mortality in seven studies (Figure 3). With the increase in fish consumption, risk estimates showed a sharp decline and then leveled off. Compared with the reference group (0 g per day), the RR reached a lowest point at ~ 60–80 g per day. The selection of different numbers and positions of knots did not significantly change these results. DISCUSSIONS To our knowledge, this is the first meta-analysis to quantitatively assess the association between fish consumption and all-cause mortality from prospective cohort studies, which indicated that a higher intake of fish was associated with about a 6% significantly lower risk of all-cause mortality. Results from the dose–response meta-analysis suggested that fish consumption was associated with all-cause mortality in a nonlinear manner with a steeper decrease in all-cause mortality at intakes below ~ 60 g per day. No publication bias was detected in our study. Findings from the present meta-analysis are consistent with previous meta-analysis on fish consumption associated with the mortality of major chronic diseases, including cardiovascular disease and cancer. A meta-analysis of 17 prospective cohort studies indicated that every 15 g per day increment of fish intake decreased the risk of CHD mortality by 6% (RR = 0.94, 95% CI: 0.90, 0.98).14 Numbers of studies found that fish consumption decreased the mortality of specific cancers, such as colorectal,40 lung41 and prostate cancers.42 It is biologically plausible that fish consumption is associated with all-cause mortality. First, n-3 PUFAs are particularly rich in fish. Mozaffarian and Rimm43 conducted a meta-analysis of 15 © 2015 Macmillan Publishers Limited
randomized controlled trials and found that marine n-3 PUFAs reduced total mortality by 17% (pooled RR = 0.83, 95% CI: 0.68, 1.00; P = 0.046). Multiple mechanisms of n-3 PUFAs are involved in this chemopreventive activity, including cell growth inhibition and enhanced apoptosis, suppression of neoplastic transformation and antiangiogenicity.16,44 In addition, it could be attributed to a wider array of nutrients that are abundant in fish, such as vitamins,45,46 essential amino acids47,48 and trace elements.49,50 Third, higher fish consumption may simply be an indicator of a healthier dietary pattern or higher socioeconomic status, which is associated with a better medical care and a lower mortality.51,52 Three studies were excluded because they provided risk estimates using moderate fish consumption (third or fourth quintile) as reference. However, we could not calculate risk estimates using the information available. In two European studies,53,54 no association was seen for higher fish consumption and all-cause mortality when compared with moderate fish intake. However, there seemed to be a U-shaped trend for fatty fish consumption on risk of total mortality.53 Another study55 was conducted in Japan and reported that a RR of 0.99 (95% CI: 0.77, 1.27) for all-cause mortality for subjects who ate fish more than twice a day compared with those who ate it only one or two times per week. With moderate significant heterogeneity found, there are some reasons to explain the inconsistency between studies. First, in sensitivity analysis, we found that the risk estimates were different between locations, which may reflect racial disparities. Second, even though the consumption of fish for almost all studies was measured by a structured FFQ, various kinds of fish and cooking methods may have different effects. Different fish have different nutritional profiles and biological effects, one obvious example European Journal of Clinical Nutrition (2015) 1 – 7
Fish consumption and all-cause mortality L-G Zhao et al
6 95%CI
Table 2.
Pooled relative risks for the highest compared with the lowest fish consumption and all-cause mortality n
RR
95% CI
I2 (%)
P-value
All studies
12
0.94
0.90, 0.98
39.1
0.055
Gender Male Female
8 6
0.93 0.94
0.85, 1.01 0.89, 1.00
55.0 21.5
0.030 0.272
Region USA Asia Europe Validated FFQ Yes No
1.00 Relative Risk
Subgroups
Spline Model
1.05
0.95
0.90
0.85
6 4 2
0.89 0.93 1.04
0.83, 0.95 0.88, 0.99 0.98, 1.11
0.0 35.7 0.0
0.795 0.169 0.836
0
20
40
60
80
100
120
140
160
Fish consumption, g/day
4 8
0.92 0.95
0.86, 0.99 0.90, 1.01
45.1 34.9
0.105 0.129
Year of publication o2008 6 2008+ 6
0.96 0.90
0.90, 1.01 0.84, 0.96
61.9 0.0
0.010 0.809
Sample size o40 000 40 000+
6 6
0.89 0.96
0.83, 0.96 0.90, 1.01
0.0 62.2
0.829 0.010
Duration of follow-up o12 year 6 12+ 6
0.91 0.97
0.86, 0.97 0.92, 1.03
40.9 25.4
0.106 0.226
Major confounders adjusted Education Yes 8 0.95 No 4 0.90
0.90, 0.99 0.83, 0.99
50.5 0.0
0.027 0.501
Body mass index Yes 10 No 2
0.94 0.90
0.90, 0.99 0.80, 1.02
43.3 0.0
0.042 0.320
Physical activity Yes No
9 3
0.93 0.96
0.88, 0.98 0.87, 1.05
40.2 51.8
0.073 0.101
Energy Yes No
9 3
0.95 0.88
0.91, 0.99 0.79, 0.97
45.8 0.0
0.041 0.549
Red meat intake Yes No
4 8
0.95 0.93
0.86, 1.05 0.89,0.96
66.0 0.0
0.019 0.509
Smoking and drinking Yes 11 No 1
0.94 0.98
0.90, 0.98 0.80, 1.21
42.9 —
0.040 —
Fruit and vegetables intake Yes 8 0.95 No 4 0.90
0.89, 1.00 0.83, 0.98
59.3 0.0
0.008 0.938
Abbreviations: CI, confidence interval; FFQ, food frequency questionnaire; RR, relative risk. I2, measure of heterogeneity; P, P-value for heterogeneity.
being white fish and fatty fish. This inaccurate classification of fish may lead to potential poor allocation of nutrient composition to food, which may attenuate the associations. Furthermore, the mortality of population from different country varied considerably, and these differences between each study may reflect a different constituent ratio of cause from cardiovascular diseases, cancers or other diseases. The present study has a number of strengths, such as large sample sizes (57 641 deaths among 672 389 individuals), follow-up European Journal of Clinical Nutrition (2015) 1 – 7
Figure 3. Relative risks of all-cause mortality associated with fish consumption (from seven studies). Fish consumption was modeled with restricted cubic splines in a fixed-effects dose–response model. The P-values for nonlinearity were 0.020 for fish consumption. Compared with the reference group (0 g per day), the RR is 0.94 (95% CI: 0.92, 0.97), 0.90 (95% CI: 0.86, 0.95), 0.88 (95% CI: 0.83, 0.93), 0.88 (95% CI: 0.84, 0.92), 0.89 (95% CI: 0.84, 0.94) and 0.89 (95% CI: 0.83, 0.96) for 20, 40, 60, 80, 100 and 120 g per day, respectively. RR, relative risks; CI, confidence interval.
duration of at least 5 years and prospective cohort design. Furthermore, our findings were stable and robust in subgroups and sensitivity analyses. Our study also has several limitations. First, the pooled results may be biased by residual confounding that is inherent in the original studies.56 Second, although our major interest was all-cause mortality, we ought to further infer which causes of death were responsible for the associations we observed.57 Effects on total mortality in a population would therefore depend on the proportion of deaths due to CHD. However, data are lacking from the retrieved studies to conduct a subgroup analysis of causespecific mortality. Third, we have not performed subgroup analysis according to fish type (fresh fish and processed fish, fatty fish and non-fatty fish), because of the lack of information available from original studies. Finally, because of this meta-analysis only including published studies, the conclusion should be made with caution. In conclusion, there is an inverse association between fish consumption and all-cause mortality. According to the dose– response analysis, intake of at least 60 g per day fish should be recommended. Potential public health benefits may exist based on the observation that increased dietary fish intake was associated with a decreased risk of all-cause mortality. CONFLICT OF INTEREST The authors declare no conflict of interest.
ACKNOWLEDGEMENTS We thank the original studies for the contribution to conduct our meta-analysis. This study was supported by the fund of Shanghai Health Bureau Key Disciplines and Specialties Foundation. Y-BX obtained the funding, conducted the research design and also had primary responsibility for the final content; L-GZ, J-WS and YY analyzed the data and interpreted the results; L-GZ drafted first manuscript; all authors critically reviewed and approved the manuscript.
REFERENCES 1 Ezzati M, Riboli E. Can noncommunicable diseases be prevented? Lessons from studies of populations and individuals. Science 2012; 337: 1482–1487. 2 Mozaffarian D, Appel LJ, Van Horn L. Components of a cardioprotective diet: new insights. Circulation 2011; 123: 2870–2891.
© 2015 Macmillan Publishers Limited
Fish consumption and all-cause mortality L-G Zhao et al
7 3 Li YH, Zhou CH, Pei HJ, Zhou XL, Li LH, Wu YJ et al. Fish consumption and incidence of heart failure: a meta-analysis of prospective cohort studies. Chin Med J (Engl) 2013; 126: 942–948. 4 Konig A, Bouzan C, Cohen JT, Connor WE, Kris-Etherton PM, Gray GM et al. A quantitative analysis of fish consumption and coronary heart disease mortality. Am J Prev Med 2005; 29: 335–346. 5 Whelton SP, He J, Whelton PK, Muntner P. Meta-analysis of observational studies on fish intake and coronary heart disease. Am J Cardiol 2004; 93: 1119–1123. 6 Larsson SC, Orsini N. Fish consumption and the risk of stroke: a dose-response meta-analysis. Stroke 2011; 42: 3621–3623. 7 Zhang M, Picard-Deland E, Marette A. Fish and marine omega-3 polyunsatured fatty acid consumption and incidence of type 2 diabetes: a systematic review and meta-analysis. Int J Endocrinol 2013; 2013: 501015. 8 Song J, Su H, Wang BL, Zhou YY, Guo LL. Fish consumption and lung cancer risk: systematic review and meta-analysis. Nutr Cancer 2014; 66: 539–549. 9 Luo J, Yang Y, Liu J, Lu K, Tang Z, Liu P et al. Systematic review with meta-analysis: meat consumption and the risk of hepatocellular carcinoma. Aliment Pharmacol Ther 2014; 39: 913–922. 10 Salehi M, Moradi-Lakeh M, Salehi MH, Nojomi M, Kolahdooz F. Meat, fish, and esophageal cancer risk: a systematic review and dose-response meta-analysis. Nutr Rev 2013; 71: 257–267. 11 Wu S, Feng B, Li K, Zhu X, Liang S, Liu X et al. Fish consumption and colorectal cancer risk in humans: a systematic review and meta-analysis. Am J Med 2012; 125: 551–559. 12 Kolahdooz F, van der Pols JC, Bain CJ, Marks GC, Hughes MC, Whiteman DC et al. Meat, fish, and ovarian cancer risk: results from 2 Australian case-control studies, a systematic review, and meta-analysis. Am J Clin Nutr 2010; 91: 1752–1763. 13 Lee JE, McLerran DF, Rolland B, Chen Y, Grant EJ, Vedanthan R et al. Meat intake and cause-specific mortality: a pooled analysis of asian prospective cohort studies. Am J Clin Nutr 2013; 98: 1032–1041. 14 Zheng J, Huang T, Yu Y, Hu X, Yang B, Li D. Fish consumption and CHD mortality: an updated meta-analysis of seventeen cohort studies. Public Health Nutr 2012; 15: 725–737. 15 Bell GA, Kantor ED, Lampe JW, Kristal AR, Heckbert SR, White E. Intake of long-chain omega-3 fatty acids from diet and supplements in relation to mortality. Am J Epidemiol 2014; 179: 710–720. 16 Chapkin RS, Davidson LA, Ly L, Weeks BR, Lupton JR, McMurray DN. Immunomodulatory effects of (n-3) fatty acids: Putative link to inflammation and colon cancer. J Nutr 2007; 137: 200S–204S. 17 Saravanan P, Davidson NC, Schmidt EB, Calder PC. Cardiovascular effects of marine omega-3 fatty acids. Lancet 2010; 376: 540–550. 18 Nanri A, Mizoue T, Noda M, Takahashi Y, Matsushita Y, Poudel-Tandukar K et al. Fish intake and type 2 diabetes in Japanese men and women: The Japan Public Health Center-based Prospective Study. Am J Clin Nutr 2011; 94: 884–891. 19 Kappeler R, Eichholzer M, Rohrmann S. Meat consumption and diet quality and mortality in NHANES III. Eur J Clin Nutr 2013; 67: 598–606. 20 Takata Y, Zhang X, Li H, Gao YT, Yang G, Gao J et al. Fish intake and risks of total and cause-specific mortality in 2 population-based cohort studies of 134,296 men and women. Am J Epidemiol 2013; 178: 46–57. 21 Olsen A, Egeberg R, Halkjaer J, Christensen J, Overvad K, Tjonneland A. Healthy aspects of the Nordic diet are related to lower total mortality. J Nutr 2011; 141: 639–644. 22 Yamagishi K, Iso H, Date C, Fukui M, Wakai K, Kikuchi S et al. Fish, omega-3 polyunsaturated fatty acids, and mortality from cardiovascular diseases in a nationwide community-based cohort of Japanese men and women the JACC (Japan collaborative cohort study for evaluation of cancer risk) study. J Am Coll Cardiol 2008; 52: 988–996. 23 Ness AR, Maynard M, Frankel S, Smith GD, Frobisher C, Leary SD et al. Diet in childhood and adult cardiovascular and all cause mortality: the boyd orr cohort. Heart 2005; 91: 894–898. 24 Folsom AR, Demissie Z. Fish intake, marine omega-3 fatty acids, and mortality in a cohort of postmenopausal women. Am J Epidemiol 2004; 160: 1005–1010. 25 Nagata C, Takatsuka N, Shimizu H. Soy and fish oil intake and mortality in a japanese community. Am J Epidemiol 2002; 156: 824–831. 26 Gillum RF, Mussolino M, Madans JH. The relation between fish consumption, death from all causes, and incidence of coronary heart disease. The NHANES i epidemiologic follow-up study. J Clin Epidemiol 2000; 53: 237–244. 27 Albert CM, Hennekens CH, O'Donnell CJ, Ajani UA, Carey VJ, Willett WC et al. Fish consumption and risk of sudden cardiac death. JAMA 1998; 279: 23–28. 28 Daviglus ML, Stamler J, Orencia AJ, Dyer AR, Liu K, Greenland P et al. Fish consumption and the 30-year risk of fatal myocardial infarction. N Engl J Med 1997; 336: 1046–1053.
© 2015 Macmillan Publishers Limited
29 Schisterman EF, Cole SR, Platt RW. Overadjustment bias and unnecessary adjustment in epidemiologic studies. Epidemiology 2009; 20: 488–495. 30 Aune D, Greenwood DC, Chan DSM, Vieira R, Vieira AR, Navarro Rosenblatt DA et al. Body mass index, abdominal fatness and pancreatic cancer risk: a systematic review and non-linear dose-response meta-analysis of prospective studies. Ann Oncol 2012; 23: 843–852. 31 DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986; 7: 177–188. 32 Jackson D, White IR, Thompson SG. Extending DerSimonian and Laird's methodology to perform multivariate random effects meta‐analyses. Stat Med 2010; 29: 1282–1297. 33 Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992; 135: 1301–1309. 34 Orsini N, Bellocco R, Greenland S. Generalized least squares for trend estimation of summarized dose-response data. Stata Journal 2006; 6: 40–57. 35 Liu Q, Cook NR, Bergstrom A, Hsieh C. A two-stage hierarchical regression model for meta-analysis of epidemiologic nonlinear dose-response data. Comput Stat Data Anal 2009; 53: 4157–4167. 36 Orsini N, Li R, Wolk A, Khudyakov P, Spiegelman D. Meta-analysis for linear and nonlinear dose-response relations: examples, an evaluation of approximations, and software. Am J Epidemiol 2012; 175: 66–73. 37 Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ 2003; 327: 557–560. 38 Egger M, Davey SG, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997; 315: 629–634. 39 Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publication bias. Biometrics 1994; 50: 1088–1101. 40 Caygill CP, Hill MJ. Fish n-3 fatty acids and human colorectal and breast cancer mortality. Eur J Cancer Prev 1995; 4: 329–332. 41 Zhang J, Temme EH, Kesteloot H. Fish consumption is inversely associated with male lung cancer mortality in countries with high levels of cigarette smoking or animal fat consumption. Int J Epidemiol 2000; 29: 615–621. 42 Szymanski KM, Wheeler DC, Mucci LA. Fish consumption and prostate cancer risk: a review and meta-analysis. Am J Clin Nutr 2010; 92: 1223–1233. 43 Mozaffarian D, Rimm EB. Fish intake, contaminants, and human health: Evaluating the risks and the benefits. JAMA 2006; 296: 1885–1899. 44 Tavani A, Pelucchi C, Parpinel M, Negri E, Franceschi S, Levi F et al. N-3 polyunsaturated fatty acid intake and cancer risk in Italy and Switzerland. Int J Cancer 2003; 105: 113–116. 45 Pilz S, Tomaschitz A, Drechsler C, Zittermann A, Dekker JM, Marz W. Vitamin D supplementation: a promising approach for the prevention and treatment of strokes. Curr Drug Targets 2011; 12: 88–96. 46 He K, Merchant A, Rimm EB, Rosner BA, Stampfer MJ, Willett WC et al. Folate, vitamin B-6, and B-12 intakes in relation to risk of stroke among men. Stroke 2004; 35: 169–174. 47 Militante JD, Lombardini JB. Treatment of hypertension with oral taurine: experimental and clinical studies. Amino Acids 2002; 23: 381–393. 48 Moncada S, Higgs A. The L-arginine-nitric oxide pathway. N Engl J Med 1993; 329: 2002–2012. 49 D'Elia L, Barba G, Cappuccio FP, Strazzullo P. Potassium intake, stroke, and cardiovascular disease a meta-analysis of prospective studies. J Am Coll Cardiol 2011; 57: 1210–1219. 50 Hoption CS. Hypothesis: dietary iodine intake in the etiology of cardiovascular disease. J Am Coll Nutr 2006; 25: 1–11. 51 He K, Rimm EB, Merchant A, Rosner BA, Stampfer MJ, Willett WC et al. Fish consumption and risk of stroke in men. JAMA 2002; 288: 3130–3136. 52 Hu FB, Bronner L, Willett WC, Stampfer MJ, Rexrode KM, Albert CM et al. Fish and omega-3 fatty acid intake and risk of coronary heart disease in women. JAMA 2002; 287: 1815–1821. 53 Engeset D, Braaten T, Teucher B, Kühn T, Bueno-de-Mesquita HB, Leenders M et al. Fish consumption and mortality in the European Prospective Investigation into Cancer and Nutrition cohort. Eur J Epidemiol 2014; 30: 57–70. 54 Osler M, Andreasen AH, Hoidrup S. No inverse association between fish consumption and risk of death from all-causes, and incidence of coronary heart disease in middle-aged, Danish adults. J Clin Epidemiol 2003; 56: 274–279. 55 Nakamura Y, Ueshima H, Okamura T, Kadowaki T, Hayakawa T, Kita Y et al. Association between fish consumption and all-cause and cause-specific mortality in Japan: NIPPON DATA80, 1980-99. Am J Med 2005; 118: 239–245. 56 Egger M, Schneider M, Davey SG. Spurious precision? Meta-analysis of observational studies. BMJ 1998; 316: 140–144. 57 Weiss NS. All-cause mortality as an outcome in epidemiologic studies: proceed with caution. Eur J Epidemiol 2014; 29: 147–149.
European Journal of Clinical Nutrition (2015) 1 – 7
ORIGINAL ARTICLE
Fish consumption and all-cause mortality: a meta-analysis of cohort studies L-G Zhao1,2, J-W Sun1,2, Y Yang1,2, X Ma1,2, Y-Y Wang1,2 and Y-B Xiang1,2 BACKGROUND/OBJECTIVES: Although fish consumption may have an influence on specific mortality of major chronic diseases, the relationship between fish consumption and all-cause mortality remains inconsistent. SUBJECTS/METHODS: We performed a systematic search of publications using PubMed and Web of science up to 31 December 2014. Summary relative risk (RR) for the highest versus lowest category of fish consumption on risk of all-cause mortality was calculated by using a random effects model. Potential nonlinear relation was tested by modeling fish intake using restricted cubic splines with three knots at fixed percentiles of the distribution. RESULTS: Twelve prospective cohort studies with 672 389 participants and 57 641 deaths were included in this meta-analysis. Compared with the lowest category, the highest category of fish intake was associated with about a 6% significantly lower risk of all-cause mortality (RR = 0.94, 95% confidence interval (CI): 0.90, 0.98; I2 = 39.1%, P = 0.06). The dose–response analysis indicated a nonlinear relationship between fish consumption and all-cause mortality. Compared with never consumers, consumption of 60 g of fish per day was associated with a 12% reduction (RR = 0.88, 95% CI: 0.83, 0.93) in risk of total death. CONCLUSIONS: These results imply that fish consumption was associated with a reduced risk of all-cause mortality. European Journal of Clinical Nutrition advance online publication, 13 May 2015; doi:10.1038/ejcn.2015.72
INTRODUCTION Noncommunicable diseases, such as cancers, cardiovascular diseases and type 2 diabetes, are responsible for about two-thirds of deaths worldwide.1 Therefore, it is urgent to develop efficient strategies that can prevent noncommunicable diseases by reducing their major risk factors and recommending healthy lifestyles, including a well-balanced diet. Fish, as a widely consumed food of human diet, has many health benefits.2 A growing body of evidence indicates that high consumption of fish may decrease the risk of chronic diseases like coronary heart disease (CHD),3–5 stroke,6 type 2 diabetes7 and certain cancers8–12 and, consequently, also for all-cause mortality.13 A systematic review of 17 cohort studies showed that fish consumption had significantly beneficial effects on the prevention of CHD mortality.14 Recently, the Vitamins and Lifestyle Study (VITAL Study) reported that higher intake of fish was significantly associated with a lower risk of all-cause mortality.15 Fish is the most common dietary source of long-chain n-3 polyunsaturated fatty acids (n-3 PUFAs), which has been demonstrated to have antiatherosclerotic and antithrombotic effects.16,17 However, fish types, cooking methods, toxic metals and other environmental contaminants in fish may have different roles on human health.18 In addition, evidence from published epidemiological studies did not show consistent results between fish consumption and all-cause mortality.13,15,19–28 Hence, we conducted a meta-analysis to assess this relationship quantitatively. 1
MATERIALS AND METHODS Search strategy We performed a systematic search in PubMed (www.ncbi.nlm.nih.gov/ pubmed/) and Web of Science (www.webofknowledge.com/) for studies published before 31 December 2014. We used the search terms as ‘fish intake’, ‘fish consumption’ in combination with ‘mortality’, ‘death’ or ‘fatal’. Furthermore, we manually reviewed the reference lists of identified papers, previous reviews and relevant meta-analyses to identify any studies that were not found from the preliminary literature searches. Two investigators (L-GZ and J-WS) independently conducted the literature search. The study was eligible for inclusion if it had the following: (1) had a cohort design based on general healthy population and published in English, (2) used fish consumption as the exposure of interest, (3) identified total death or all-cause mortality as the outcome of interest and (4) reported the relative risk (RR) or hazard risk (HR) (highest versus lowest) and the corresponding 95% confidence interval (CI).
Data extraction From each eligible study, we extracted the first author’s name, study population and region, study period, number of participants, gender, exposure measurement, covariates adjusted RR or HR and the corresponding 95% CI, as well as numbers of cases and person-years or persons for each category of fish consumption. We used the risk estimates that derived from the multivariable-adjusted model that did not include the potential intermediates (such as, blood glucose, triglyceride or inflammatory biomarkers) on the causal pathway from exposure to disease in original studies, because control for an intermediate variable would bring overadjustment bias.29 Data extraction was implemented independently by two authors. Any inconsistency was solved by the third author (YY). In order to perform a dose–response analysis, when fish consumption was measured as ‘servings per week’, we converted this into ‘gram per day’
State Key Laboratory of Oncogene and Related Genes, Shanghai Cancer Institute, Renji Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, China and Department of Epidemiology, Shanghai Cancer Institute, Renji Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, China. Correspondence: Professor Y-B Xiang, Department of Epidemiology, Shanghai Cancer Institute, Renji Hospital, Shanghai Jiaotong University School of Medicine, No. 25, Lane 2200, Xie Tu Road, Shanghai 200032, China. E-mail: [email protected] Received 15 August 2014; revised 16 March 2015; accepted 6 April 2015 2
Fish consumption and all-cause mortality L-G Zhao et al
2 as a standard measure of fish intake according to the information provided by the identified study.24 For each category, we assigned the median value of each fish category. If the upper bound in the highest category was not available, we assumed that it had the same amplitude as the preceding one. If the distribution of cases or person-years in the study was not reported, we estimated the distribution using the methods described by Aune et al.30
Records identified through PubMed (n=982) and Web of Science (n=2445) Records with duplicate titles /abstracts were excluded (n=822). Records after duplicates removed (n=2605) Records were excluded based on screening of titles and/or abstracts using general criteria (n=2564)
Statistical analysis We conducted two types of meta-analyses. First, we combined the RRs for the highest versus lowest category of fish consumption on risk of all-cause mortality by using the DerSimonian and Laird random-effects model, which incorporated both within- and between-study variability.31,32 Second, for the dose–response meta-analysis, the method proposed by Greenland and Longnecker33 and Orsini et al.34 was used to calculate the trend from the correlated log-relative risks across categories of fish consumption. In order to avoid bias estimates, we centered each study to the baseline reference of fish intake.35 Potential nonlinear relation was tested by modeling fish intake using restricted cubic splines with three knots at fixed percentiles (10, 50 and 90%) of the distribution.32,34 A P-value for nonlinearity of the meta-analysis was evaluated by testing the null hypothesis that the coefficient of the second spline was equal to zero.36 Heterogeneity among studies was evaluated by using the Q and I2 statistics. I2 took values between 0 and 100% and there may be important heterogeneity when I2450% or P for Q statistic o0.10.37 In a sensitivity analysis, we sequentially omitted one study at a time from the metaanalysis to examine whether the main results were influenced by a particular study. In order to test the robustness of the combined estimates, separate meta-analysis was performed stratified by gender (men or women), region (USA, Asia or Europe), methods of dietary assessment (validated food frequency questionnaire (FFQ) or others), sample size (using 40 000 participants as the cut point), follow-up duration (using 12 years as the cut point), year of publication (before 2008 or after 2008) and important confounding adjustments (yes or no) for education, body mass index, physical activity, intake of fruit and vegetables, red meat and total energy. Small study bias (for example, publication bias) was assessed through visual inspection of funnel plots and using Egger’s38 and Begg’s tests.39 P o 0.10 was deemed to possess publication bias. All statistical analyses were conducted using STATA software (version 12.0; StataCorp, College Station, TX, USA). P-values with two side of o0.05 were considered statistically significant if not specified.
RESULTS Literature search and study characteristics We obtained 2605 articles after removal of duplicates from our preliminary search. After screening the title and abstract, 2564 records were excluded using the general criteria, such as review, animal research and retrospective study. Forty articles were potentially related to our issue for further scrutiny, of which 29 articles were excluded according to the inclusion criteria. With one study19 obtained from checking reference lists, thus twelve studies13,15,19–28 were included in the meta-analysis (Figure 1). Takata et al.20 reported the results of Shanghai Women Health Study (SWHS) and the Shanghai Men Health Study (SMHS), whereas the SWHS was updated by a pooled cohort study13 including eight cohorts in Asia. Then, we only abstracted the data of the SMHS from this study20 in meta-analysis of highest versus lowest category. Of the twelve identified studies (Table 1), the median follow-up time was 12 years. Combined, these studies included 672 389 cohort members and 57 641 deaths. Six studies15,19,24,26–28 were conducted in the United States, four13,20,22,25 in Asia and two21,23 in Europe. Most studies13,15,19–22,24–27 used a structured FFQ to estimate fish intake. All studies adjusted for age and gender as potential confounders (if applicable). Only one study23 failed to adjust for smoking status and alcohol drinking. European Journal of Clinical Nutrition (2015) 1 – 7
Records obtained from titles/abstracts screening (n=41)
Records obtained from full-text screening (n=11)
Records were excluded as: (1) No RRs/HRs or 95%CIs for highest/lowest were reported (n=9); (2) Dietary pattern rich in fish or sea food as exposure is reported (n=11); (3) The outcome is specific disease mortality (n=5); (4) The base population are patients (n=5). Records obtained from checking reference lists of retrieved articles (n=1).
Studies included in meta-analysis (n=12) (1) For highest/lowest analysis (n=12); (2) For dose-response analysis (n=7).
Figure 1. Flow diagram for selection of studies in meta-analysis of fish consumption and all-cause mortality. RR, relative risk; HR, hazard ratio; CI, confidence interval.
Highest versus lowest category Overall, analysis of 12 prospective studies significantly showed a 6% reduction in risk of all-cause mortality with high intake of fish (RR = 0.94, 95% CI: 0.90, 0.98) with moderate evidence of betweenstudy heterogeneity (I2 = 39.1%, P = 0.06) using a random-effects model (Figure 2). Visual inspection of Begg’s test and Egger’s test suggested that there was no publication bias (Begg’s test: P = 0.26; Egger’s test: P = 0.14). Sensitivity analysis and subgroup analysis In sensitivity analysis, we recalculated the combined results by sequentially excluding each study to examine the influence of individual study on the pooled estimate. The risk estimates ranged from 0.92 (95% CI: 0.89, 0.95; I2 = 0.0%, P = 0.462), after excluding Olsen,21 to 0.95 (95% CI: 0.90, 0.99; I2 = 37.4%, P = 0.072), after excluding Bell.15 None of the studies considerably affected the summary results. Subgroup analysis was performed across a number of key study characteristics (Table 2). The pooled RRs did not materially differ between genders. A lower risk was observed in the United States, whereas no association was found in Europe. There was a significantly inverse association in studies published after 2008. Stratifications by other characteristics did not have a substantial impact on the main results. Dose–response meta-analysis Although we had contacted with the authors for more detailed information, five studies were excluded from dose–response analysis because only two categories of fish consumption were considered,21 or no information was provided regarding the amount of fish intake in each category.13,15,19,26 © 2015 Macmillan Publishers Limited
© 2015 Macmillan Publishers Limited
Japan
UK
Yamagishi22
Ness23
6
7
China
Takata20
4
Denmark
Asia
Lee13
3
Olsen21
USA
Kappeler19
2
5
USA
Bell15
1
37
12.7
12
5.6
6.6–15.6
22
5
Cohort size
British
Japanese
Danes
Chinese
NA
7008
4126 (2383 men and 1743 women)
2170
24 283 (14 326 men and 9957 women)
3683 (1908 men and 1775 women)
3051
No. of deaths
4028 (1995 1010 men and 2033 women)
57 972 (22 881 men and 35 091 women
50 290 (23 274 men and 27 016 women)
73 159
296 721 (112 310 men and 184 411 women)
Mixed: main 17 611 non-hispanic white (76.5%)
Mixed: main 70 495 white (93.3%)
Follow-up Populations years
The food records
FFQ with 33 items
FFQ with 192 items
FFQ
FFQ
FFQ with 81 items
FFQ with 120 items
Dietary assessment method
General characteristics of prospective studies of fish consumption and all-cause mortality
ID First author Country (reference no.)
Table 1.
44.5 versus 1.8 g per day
Men: 86 versus o19 g per day. Women: ⩾ 84 versus o19 g per day
Men: ⩾ 41 versus o41 g per day. Women: ⩾ 35 versus o35 g per day
121.9 versus 11.1 g per day
Both
Both
Both
Men
Both
Age, sex, energy, childhood family food expenditure, father’s social class, district of residence as a child, period of birth, season when studied as a child and Townsend score for current address or place of death.
Age, gender, energy, history of hypertension and diabetes mellitus, smoking status, alcohol consumption, BMI, mental stress, walking, sports, education levels, total energy, and dietary intakes of cholesterol, saturated and n-6 polyunsaturated fatty acids, vegetables and fruit.
Age, time under study, smoking status, smoking duration, current tobacco consumption, time since cessation, alcohol intake, school education, participation in sports, time spent on sports per week, BMI, intake of red meat, intake of processed meat intake, total energy intake and other dietary components.
Age at baseline, total energy intake, income, occupation, education, comorbidity index, physical activity level, red meat intake, poultry intake, total vegetable intake, total fruit intake, smoking history and alcohol consumption.
Age, BMI, education, smoking habit, rural/urban residence, alcohol intake, fruit and vegetable intake and total energy intake.
Age, race, sex, cigarette smoking, alcohol consumption, physical activity, socioeconomic status, BMI, marital status, fruit and vegetables intake, history of hypertension, diabetes, hypercholesterolemia, use of aspirin and ibuprofen, use of mineral and vitamin supplements, family history of diabetes or hypercholesterolemia; hormone replacement therapy and oral contraceptive use (in women).
49 versus 0 times per Both month
Q4 versus Q1: mean = 36.6 6 ± 37.2 g per day
Age, sex, race/ethnicity, marital status, education, total energy intake, BMI at age 45 years, average alcohol intake at age 45 years, average physical activity in the 10 years before baseline, self-rated health, mammogram in the last 2 years, prostate-specific antigen test in the last 2 years, sigmoidoscopy in the last 10 years, current use of cholesterol-lowering medication, aspirin use in the past 10 years, use of nonaspirin nonsteroidal anti-inflammatory drugs in the past 10 years, smoking, morbidity score, percentage of calories derived from trans fat, percentage of calories derived from saturated fat, number of servings per day of fruits, number of servings per day of vegetables, years of estrogen therapy, years of estrogen+progestin therapy, age at menopause, age at death of father, and age at death of mother.
Both
442 versus 0 times per year
Adjustment
Sex
Quantity
Fish consumption and all-cause mortality L-G Zhao et al
3
European Journal of Clinical Nutrition (2015) 1 – 7
(Continued )
European Journal of Clinical Nutrition (2015) 1 – 7
USA
USA
11 Albert27
12 Daviglus28
25.8
11
18.8
6.9
10.5
29 079 (13 355 men and 15 724 women)
41 836
Cohort size
20 551
Chicago men 1822
US male physicians
Mixed: main 8825 white (84.1%)
Japanese
USA postmenopausal women
Follow-up Populations years
Abbreviations: BMI, body mass index; FFQ, food frequency questionnaire.
USA
Japan
Nagata25
9
10 Gillum26
USA
Folsom24
8
ID First author Country (reference no.)
Table 1.
1042
1652
2901 (1513 men and 1388 women)
2062 (1163 men and 899 women)
4653
No. of deaths
Men
⩾ 5 times per week versus ⩽ 1 time per month
Men
Baseline age, smoking, history of diabetes, education, high school graduate, systolic blood pressure, serum cholesterol concentration, BMI, alcohol intake and physical activity.
Both
41 times per week versus never
Baseline age, education, religion, systolic pressure, serum cholesterol, number of cigarettes smoked per day, BMI, presence or absence of diabetes, presence or absence of electrocardiographic abnormalities, daily intake of energy, cholesterol, saturated, monounsaturated, fatty acids, total protein, carbohydrate, alcohol, iron, thiamine, riboflavin, niacin, vitamin C, beta carotene and retinol.
Age, aspirin and beta carotene treatment assignment, evidence of cardiovascular disease before 12-month questionnaire, BMI, smoking status, history of diabetes, history of hypertension, history of hypercholesterolemia, alcohol consumption, vigorous exercise, and vitamin E, vitamin C, multivitamin use, other dietary factors (red meat, vegetables, fruits, dairy, chicken or turkey and fried foods).
For men: age, total energy, marital status, BMI, smoking status, alcohol intake, coffee intake, exercise, and history of hypertension and diabetes mellitus. For women: age, total energy, marital status, years of education, BMI, smoking status, alcohol intake, age at menarche, menopausal status, exercise and history of diabetes mellitus.
Men: 157.8 versus Both 46.2 g per day. Women: 122.4 versus 36.6 g per day
Women Age, energy intake, educational level, physical activity level, alcohol consumption, smoking status, pack-years of cigarette smoking, age at first livebirth, estrogen use, vitamin use, BMI, waist/hip ratio, diabetes, hypertension, intake of whole grains, fruit and vegetables, red meat, cholesterol and saturated fat.
42.5 versus o0.5 serving per day
Adjustment
Sex
Quantity
Standard ⩾35 versus 0 g questionnaires per day
FFQ
3-month FFQ
FFQ with 169 items
FFQ with 127 items
Dietary assessment method
Fish consumption and all-cause mortality L-G Zhao et al
4
© 2015 Macmillan Publishers Limited
Fish consumption and all-cause mortality L-G Zhao et al
5 study
year
sex
quantity
Bell et al.
2014
Both
>42 times/y vs. 0 times/y
0.86 (0.76, 0.98)
Kappeler et al.
2013
Both
>9 times/m vs. 0 times/m
0.93 (0.78, 1.11)
Lee et al.
2013
Men
Q4 vs. Q1: mean=36.6 g/d
1.05 (0.95, 1.16)
Lee et al.
2013
Women Q4 vs. Q1: mean=36.6 g/d
0.91 (0.85, 0.97)
Takata et al.
2013
Men
121.9 g/d vs. 11.1 g/d
0.86 (0.74, 1.00)
Olsen et al.
2011
Men
>=41 vs. =35 vs. =86 vs. =84 vs. 2.5 vs. 1 times/w vs. never
0.89 (0.73, 1.09)
Gillum et al.
2000
Women >1 times/w vs. never
0.88 (0.71, 1.09)
Albert et al.
1998
Men
>=5 times/w vs. =35 g/d vs. 0 g/d
0.85 (0.64, 1.10)
RR (95% CI)
Overall (I-squared = 39.1%, p = 0.055)
0.94 (0.90, 0.98)
NOTE: Weights are from random effects analysis
0.5
0.6
0.7
0.8
0.9
1
1.2
Figure 2. Relative risk of all-cause mortality for highest versus lowest category of fish intake. Overall relative risk calculated with random effects model.
We observed a significantly nonlinear relationship (P = 0.020) between fish consumption and all-cause mortality in seven studies (Figure 3). With the increase in fish consumption, risk estimates showed a sharp decline and then leveled off. Compared with the reference group (0 g per day), the RR reached a lowest point at ~ 60–80 g per day. The selection of different numbers and positions of knots did not significantly change these results. DISCUSSIONS To our knowledge, this is the first meta-analysis to quantitatively assess the association between fish consumption and all-cause mortality from prospective cohort studies, which indicated that a higher intake of fish was associated with about a 6% significantly lower risk of all-cause mortality. Results from the dose–response meta-analysis suggested that fish consumption was associated with all-cause mortality in a nonlinear manner with a steeper decrease in all-cause mortality at intakes below ~ 60 g per day. No publication bias was detected in our study. Findings from the present meta-analysis are consistent with previous meta-analysis on fish consumption associated with the mortality of major chronic diseases, including cardiovascular disease and cancer. A meta-analysis of 17 prospective cohort studies indicated that every 15 g per day increment of fish intake decreased the risk of CHD mortality by 6% (RR = 0.94, 95% CI: 0.90, 0.98).14 Numbers of studies found that fish consumption decreased the mortality of specific cancers, such as colorectal,40 lung41 and prostate cancers.42 It is biologically plausible that fish consumption is associated with all-cause mortality. First, n-3 PUFAs are particularly rich in fish. Mozaffarian and Rimm43 conducted a meta-analysis of 15 © 2015 Macmillan Publishers Limited
randomized controlled trials and found that marine n-3 PUFAs reduced total mortality by 17% (pooled RR = 0.83, 95% CI: 0.68, 1.00; P = 0.046). Multiple mechanisms of n-3 PUFAs are involved in this chemopreventive activity, including cell growth inhibition and enhanced apoptosis, suppression of neoplastic transformation and antiangiogenicity.16,44 In addition, it could be attributed to a wider array of nutrients that are abundant in fish, such as vitamins,45,46 essential amino acids47,48 and trace elements.49,50 Third, higher fish consumption may simply be an indicator of a healthier dietary pattern or higher socioeconomic status, which is associated with a better medical care and a lower mortality.51,52 Three studies were excluded because they provided risk estimates using moderate fish consumption (third or fourth quintile) as reference. However, we could not calculate risk estimates using the information available. In two European studies,53,54 no association was seen for higher fish consumption and all-cause mortality when compared with moderate fish intake. However, there seemed to be a U-shaped trend for fatty fish consumption on risk of total mortality.53 Another study55 was conducted in Japan and reported that a RR of 0.99 (95% CI: 0.77, 1.27) for all-cause mortality for subjects who ate fish more than twice a day compared with those who ate it only one or two times per week. With moderate significant heterogeneity found, there are some reasons to explain the inconsistency between studies. First, in sensitivity analysis, we found that the risk estimates were different between locations, which may reflect racial disparities. Second, even though the consumption of fish for almost all studies was measured by a structured FFQ, various kinds of fish and cooking methods may have different effects. Different fish have different nutritional profiles and biological effects, one obvious example European Journal of Clinical Nutrition (2015) 1 – 7
Fish consumption and all-cause mortality L-G Zhao et al
6 95%CI
Table 2.
Pooled relative risks for the highest compared with the lowest fish consumption and all-cause mortality n
RR
95% CI
I2 (%)
P-value
All studies
12
0.94
0.90, 0.98
39.1
0.055
Gender Male Female
8 6
0.93 0.94
0.85, 1.01 0.89, 1.00
55.0 21.5
0.030 0.272
Region USA Asia Europe Validated FFQ Yes No
1.00 Relative Risk
Subgroups
Spline Model
1.05
0.95
0.90
0.85
6 4 2
0.89 0.93 1.04
0.83, 0.95 0.88, 0.99 0.98, 1.11
0.0 35.7 0.0
0.795 0.169 0.836
0
20
40
60
80
100
120
140
160
Fish consumption, g/day
4 8
0.92 0.95
0.86, 0.99 0.90, 1.01
45.1 34.9
0.105 0.129
Year of publication o2008 6 2008+ 6
0.96 0.90
0.90, 1.01 0.84, 0.96
61.9 0.0
0.010 0.809
Sample size o40 000 40 000+
6 6
0.89 0.96
0.83, 0.96 0.90, 1.01
0.0 62.2
0.829 0.010
Duration of follow-up o12 year 6 12+ 6
0.91 0.97
0.86, 0.97 0.92, 1.03
40.9 25.4
0.106 0.226
Major confounders adjusted Education Yes 8 0.95 No 4 0.90
0.90, 0.99 0.83, 0.99
50.5 0.0
0.027 0.501
Body mass index Yes 10 No 2
0.94 0.90
0.90, 0.99 0.80, 1.02
43.3 0.0
0.042 0.320
Physical activity Yes No
9 3
0.93 0.96
0.88, 0.98 0.87, 1.05
40.2 51.8
0.073 0.101
Energy Yes No
9 3
0.95 0.88
0.91, 0.99 0.79, 0.97
45.8 0.0
0.041 0.549
Red meat intake Yes No
4 8
0.95 0.93
0.86, 1.05 0.89,0.96
66.0 0.0
0.019 0.509
Smoking and drinking Yes 11 No 1
0.94 0.98
0.90, 0.98 0.80, 1.21
42.9 —
0.040 —
Fruit and vegetables intake Yes 8 0.95 No 4 0.90
0.89, 1.00 0.83, 0.98
59.3 0.0
0.008 0.938
Abbreviations: CI, confidence interval; FFQ, food frequency questionnaire; RR, relative risk. I2, measure of heterogeneity; P, P-value for heterogeneity.
being white fish and fatty fish. This inaccurate classification of fish may lead to potential poor allocation of nutrient composition to food, which may attenuate the associations. Furthermore, the mortality of population from different country varied considerably, and these differences between each study may reflect a different constituent ratio of cause from cardiovascular diseases, cancers or other diseases. The present study has a number of strengths, such as large sample sizes (57 641 deaths among 672 389 individuals), follow-up European Journal of Clinical Nutrition (2015) 1 – 7
Figure 3. Relative risks of all-cause mortality associated with fish consumption (from seven studies). Fish consumption was modeled with restricted cubic splines in a fixed-effects dose–response model. The P-values for nonlinearity were 0.020 for fish consumption. Compared with the reference group (0 g per day), the RR is 0.94 (95% CI: 0.92, 0.97), 0.90 (95% CI: 0.86, 0.95), 0.88 (95% CI: 0.83, 0.93), 0.88 (95% CI: 0.84, 0.92), 0.89 (95% CI: 0.84, 0.94) and 0.89 (95% CI: 0.83, 0.96) for 20, 40, 60, 80, 100 and 120 g per day, respectively. RR, relative risks; CI, confidence interval.
duration of at least 5 years and prospective cohort design. Furthermore, our findings were stable and robust in subgroups and sensitivity analyses. Our study also has several limitations. First, the pooled results may be biased by residual confounding that is inherent in the original studies.56 Second, although our major interest was all-cause mortality, we ought to further infer which causes of death were responsible for the associations we observed.57 Effects on total mortality in a population would therefore depend on the proportion of deaths due to CHD. However, data are lacking from the retrieved studies to conduct a subgroup analysis of causespecific mortality. Third, we have not performed subgroup analysis according to fish type (fresh fish and processed fish, fatty fish and non-fatty fish), because of the lack of information available from original studies. Finally, because of this meta-analysis only including published studies, the conclusion should be made with caution. In conclusion, there is an inverse association between fish consumption and all-cause mortality. According to the dose– response analysis, intake of at least 60 g per day fish should be recommended. Potential public health benefits may exist based on the observation that increased dietary fish intake was associated with a decreased risk of all-cause mortality. CONFLICT OF INTEREST The authors declare no conflict of interest.
ACKNOWLEDGEMENTS We thank the original studies for the contribution to conduct our meta-analysis. This study was supported by the fund of Shanghai Health Bureau Key Disciplines and Specialties Foundation. Y-BX obtained the funding, conducted the research design and also had primary responsibility for the final content; L-GZ, J-WS and YY analyzed the data and interpreted the results; L-GZ drafted first manuscript; all authors critically reviewed and approved the manuscript.
REFERENCES 1 Ezzati M, Riboli E. Can noncommunicable diseases be prevented? Lessons from studies of populations and individuals. Science 2012; 337: 1482–1487. 2 Mozaffarian D, Appel LJ, Van Horn L. Components of a cardioprotective diet: new insights. Circulation 2011; 123: 2870–2891.
© 2015 Macmillan Publishers Limited
Fish consumption and all-cause mortality L-G Zhao et al
7 3 Li YH, Zhou CH, Pei HJ, Zhou XL, Li LH, Wu YJ et al. Fish consumption and incidence of heart failure: a meta-analysis of prospective cohort studies. Chin Med J (Engl) 2013; 126: 942–948. 4 Konig A, Bouzan C, Cohen JT, Connor WE, Kris-Etherton PM, Gray GM et al. A quantitative analysis of fish consumption and coronary heart disease mortality. Am J Prev Med 2005; 29: 335–346. 5 Whelton SP, He J, Whelton PK, Muntner P. Meta-analysis of observational studies on fish intake and coronary heart disease. Am J Cardiol 2004; 93: 1119–1123. 6 Larsson SC, Orsini N. Fish consumption and the risk of stroke: a dose-response meta-analysis. Stroke 2011; 42: 3621–3623. 7 Zhang M, Picard-Deland E, Marette A. Fish and marine omega-3 polyunsatured fatty acid consumption and incidence of type 2 diabetes: a systematic review and meta-analysis. Int J Endocrinol 2013; 2013: 501015. 8 Song J, Su H, Wang BL, Zhou YY, Guo LL. Fish consumption and lung cancer risk: systematic review and meta-analysis. Nutr Cancer 2014; 66: 539–549. 9 Luo J, Yang Y, Liu J, Lu K, Tang Z, Liu P et al. Systematic review with meta-analysis: meat consumption and the risk of hepatocellular carcinoma. Aliment Pharmacol Ther 2014; 39: 913–922. 10 Salehi M, Moradi-Lakeh M, Salehi MH, Nojomi M, Kolahdooz F. Meat, fish, and esophageal cancer risk: a systematic review and dose-response meta-analysis. Nutr Rev 2013; 71: 257–267. 11 Wu S, Feng B, Li K, Zhu X, Liang S, Liu X et al. Fish consumption and colorectal cancer risk in humans: a systematic review and meta-analysis. Am J Med 2012; 125: 551–559. 12 Kolahdooz F, van der Pols JC, Bain CJ, Marks GC, Hughes MC, Whiteman DC et al. Meat, fish, and ovarian cancer risk: results from 2 Australian case-control studies, a systematic review, and meta-analysis. Am J Clin Nutr 2010; 91: 1752–1763. 13 Lee JE, McLerran DF, Rolland B, Chen Y, Grant EJ, Vedanthan R et al. Meat intake and cause-specific mortality: a pooled analysis of asian prospective cohort studies. Am J Clin Nutr 2013; 98: 1032–1041. 14 Zheng J, Huang T, Yu Y, Hu X, Yang B, Li D. Fish consumption and CHD mortality: an updated meta-analysis of seventeen cohort studies. Public Health Nutr 2012; 15: 725–737. 15 Bell GA, Kantor ED, Lampe JW, Kristal AR, Heckbert SR, White E. Intake of long-chain omega-3 fatty acids from diet and supplements in relation to mortality. Am J Epidemiol 2014; 179: 710–720. 16 Chapkin RS, Davidson LA, Ly L, Weeks BR, Lupton JR, McMurray DN. Immunomodulatory effects of (n-3) fatty acids: Putative link to inflammation and colon cancer. J Nutr 2007; 137: 200S–204S. 17 Saravanan P, Davidson NC, Schmidt EB, Calder PC. Cardiovascular effects of marine omega-3 fatty acids. Lancet 2010; 376: 540–550. 18 Nanri A, Mizoue T, Noda M, Takahashi Y, Matsushita Y, Poudel-Tandukar K et al. Fish intake and type 2 diabetes in Japanese men and women: The Japan Public Health Center-based Prospective Study. Am J Clin Nutr 2011; 94: 884–891. 19 Kappeler R, Eichholzer M, Rohrmann S. Meat consumption and diet quality and mortality in NHANES III. Eur J Clin Nutr 2013; 67: 598–606. 20 Takata Y, Zhang X, Li H, Gao YT, Yang G, Gao J et al. Fish intake and risks of total and cause-specific mortality in 2 population-based cohort studies of 134,296 men and women. Am J Epidemiol 2013; 178: 46–57. 21 Olsen A, Egeberg R, Halkjaer J, Christensen J, Overvad K, Tjonneland A. Healthy aspects of the Nordic diet are related to lower total mortality. J Nutr 2011; 141: 639–644. 22 Yamagishi K, Iso H, Date C, Fukui M, Wakai K, Kikuchi S et al. Fish, omega-3 polyunsaturated fatty acids, and mortality from cardiovascular diseases in a nationwide community-based cohort of Japanese men and women the JACC (Japan collaborative cohort study for evaluation of cancer risk) study. J Am Coll Cardiol 2008; 52: 988–996. 23 Ness AR, Maynard M, Frankel S, Smith GD, Frobisher C, Leary SD et al. Diet in childhood and adult cardiovascular and all cause mortality: the boyd orr cohort. Heart 2005; 91: 894–898. 24 Folsom AR, Demissie Z. Fish intake, marine omega-3 fatty acids, and mortality in a cohort of postmenopausal women. Am J Epidemiol 2004; 160: 1005–1010. 25 Nagata C, Takatsuka N, Shimizu H. Soy and fish oil intake and mortality in a japanese community. Am J Epidemiol 2002; 156: 824–831. 26 Gillum RF, Mussolino M, Madans JH. The relation between fish consumption, death from all causes, and incidence of coronary heart disease. The NHANES i epidemiologic follow-up study. J Clin Epidemiol 2000; 53: 237–244. 27 Albert CM, Hennekens CH, O'Donnell CJ, Ajani UA, Carey VJ, Willett WC et al. Fish consumption and risk of sudden cardiac death. JAMA 1998; 279: 23–28. 28 Daviglus ML, Stamler J, Orencia AJ, Dyer AR, Liu K, Greenland P et al. Fish consumption and the 30-year risk of fatal myocardial infarction. N Engl J Med 1997; 336: 1046–1053.
© 2015 Macmillan Publishers Limited
29 Schisterman EF, Cole SR, Platt RW. Overadjustment bias and unnecessary adjustment in epidemiologic studies. Epidemiology 2009; 20: 488–495. 30 Aune D, Greenwood DC, Chan DSM, Vieira R, Vieira AR, Navarro Rosenblatt DA et al. Body mass index, abdominal fatness and pancreatic cancer risk: a systematic review and non-linear dose-response meta-analysis of prospective studies. Ann Oncol 2012; 23: 843–852. 31 DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986; 7: 177–188. 32 Jackson D, White IR, Thompson SG. Extending DerSimonian and Laird's methodology to perform multivariate random effects meta‐analyses. Stat Med 2010; 29: 1282–1297. 33 Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992; 135: 1301–1309. 34 Orsini N, Bellocco R, Greenland S. Generalized least squares for trend estimation of summarized dose-response data. Stata Journal 2006; 6: 40–57. 35 Liu Q, Cook NR, Bergstrom A, Hsieh C. A two-stage hierarchical regression model for meta-analysis of epidemiologic nonlinear dose-response data. Comput Stat Data Anal 2009; 53: 4157–4167. 36 Orsini N, Li R, Wolk A, Khudyakov P, Spiegelman D. Meta-analysis for linear and nonlinear dose-response relations: examples, an evaluation of approximations, and software. Am J Epidemiol 2012; 175: 66–73. 37 Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ 2003; 327: 557–560. 38 Egger M, Davey SG, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997; 315: 629–634. 39 Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publication bias. Biometrics 1994; 50: 1088–1101. 40 Caygill CP, Hill MJ. Fish n-3 fatty acids and human colorectal and breast cancer mortality. Eur J Cancer Prev 1995; 4: 329–332. 41 Zhang J, Temme EH, Kesteloot H. Fish consumption is inversely associated with male lung cancer mortality in countries with high levels of cigarette smoking or animal fat consumption. Int J Epidemiol 2000; 29: 615–621. 42 Szymanski KM, Wheeler DC, Mucci LA. Fish consumption and prostate cancer risk: a review and meta-analysis. Am J Clin Nutr 2010; 92: 1223–1233. 43 Mozaffarian D, Rimm EB. Fish intake, contaminants, and human health: Evaluating the risks and the benefits. JAMA 2006; 296: 1885–1899. 44 Tavani A, Pelucchi C, Parpinel M, Negri E, Franceschi S, Levi F et al. N-3 polyunsaturated fatty acid intake and cancer risk in Italy and Switzerland. Int J Cancer 2003; 105: 113–116. 45 Pilz S, Tomaschitz A, Drechsler C, Zittermann A, Dekker JM, Marz W. Vitamin D supplementation: a promising approach for the prevention and treatment of strokes. Curr Drug Targets 2011; 12: 88–96. 46 He K, Merchant A, Rimm EB, Rosner BA, Stampfer MJ, Willett WC et al. Folate, vitamin B-6, and B-12 intakes in relation to risk of stroke among men. Stroke 2004; 35: 169–174. 47 Militante JD, Lombardini JB. Treatment of hypertension with oral taurine: experimental and clinical studies. Amino Acids 2002; 23: 381–393. 48 Moncada S, Higgs A. The L-arginine-nitric oxide pathway. N Engl J Med 1993; 329: 2002–2012. 49 D'Elia L, Barba G, Cappuccio FP, Strazzullo P. Potassium intake, stroke, and cardiovascular disease a meta-analysis of prospective studies. J Am Coll Cardiol 2011; 57: 1210–1219. 50 Hoption CS. Hypothesis: dietary iodine intake in the etiology of cardiovascular disease. J Am Coll Nutr 2006; 25: 1–11. 51 He K, Rimm EB, Merchant A, Rosner BA, Stampfer MJ, Willett WC et al. Fish consumption and risk of stroke in men. JAMA 2002; 288: 3130–3136. 52 Hu FB, Bronner L, Willett WC, Stampfer MJ, Rexrode KM, Albert CM et al. Fish and omega-3 fatty acid intake and risk of coronary heart disease in women. JAMA 2002; 287: 1815–1821. 53 Engeset D, Braaten T, Teucher B, Kühn T, Bueno-de-Mesquita HB, Leenders M et al. Fish consumption and mortality in the European Prospective Investigation into Cancer and Nutrition cohort. Eur J Epidemiol 2014; 30: 57–70. 54 Osler M, Andreasen AH, Hoidrup S. No inverse association between fish consumption and risk of death from all-causes, and incidence of coronary heart disease in middle-aged, Danish adults. J Clin Epidemiol 2003; 56: 274–279. 55 Nakamura Y, Ueshima H, Okamura T, Kadowaki T, Hayakawa T, Kita Y et al. Association between fish consumption and all-cause and cause-specific mortality in Japan: NIPPON DATA80, 1980-99. Am J Med 2005; 118: 239–245. 56 Egger M, Schneider M, Davey SG. Spurious precision? Meta-analysis of observational studies. BMJ 1998; 316: 140–144. 57 Weiss NS. All-cause mortality as an outcome in epidemiologic studies: proceed with caution. Eur J Epidemiol 2014; 29: 147–149.
European Journal of Clinical Nutrition (2015) 1 – 7
